3D Reconstruction Based On Stereovision

ABSTRACT:3D reconstruction is one of the most popular research areas in computer graphics and computer vision. It gets three-dimensional information according to two-dimensional images. Three-dimensional space includes more information than two-dimensional space, can be accepted intuitively and globally. We can realize the virtual scene's recurrence using this technology and observe the model from any viewpoints stereoscopically and apperceive the world well.In this paper, we use 3D Reconstruction based on Stereovision and a lot of technologies, such as camera calibration, point cloud building, surface reconstruction, texture mapping and so on. Our work includes several parts.The first part is camera calibration, we use the calibration method with a calibration object, and get the camera parameters through a series of mathematical calculations. We use the calibration tool developed by Klaus Strobl.The second part is point cloud building, we use the method called PMVS. We detect the DoG and Harris features in the images and do some works like matching, expansion, filtering to get the feature pairs and then get the point cloud by triangulation.The third part is surface reconstruction, we use the method called Poisson Surface Reconstruction. Build a spatial Poisson formulation through the oriented points in the point cloud to approximate the surface, extract an appropriate isosurface based on a series of criterions to obtain the visual hull.The forth part is texture mapping, the parameterization method is applied to establish the correspondences between the visual hull and the texture image, and then fill the texture image. We make the visual hull looked more dramatic by drawing and rendering the visual hull using the texture image.In this paper, we also introduce applications of the reconstructed results in our lab's platform and demonstrate the feasibility of this reconstruction method. According to our work until now, we do some conclusions and prospects.